UCF_ViT-B.yaml

DDP_CONFIG:
  WORLD_SIZE: 1
  WORLD_RANK: 0
  GPU_WORLD_SIZE: 8
  GPU_WORLD_RANK: 0
  DIST_URL: 'tcp://{}:11591' 
  WORLD_URLS: ["{}"]
  AUTO_RANK_MATCH: True
  DIST_BACKEND: 'nccl'
  GPU: 0
  DISTRIBUTED: True

CONFIG:
  EVAL_ONLY: True
  TWO_STREAM: False
  USE_LFB: False
  USE_LOCATION: False

  VAL:
    BATCH_SIZE: 1
    FREQ: 2

  DATA:
    DATASET_NAME: 'ucf'
    ANNO_PATH: '/path/to/dataset/UCF101_v2/UCF101v2-GT.pkl'
    DATA_PATH: '/path/to/dataset/UCF101_v2/rgb-images'
    NUM_CLASSES: 24
    MULTIGRID: False
    IMG_SIZE: 224
    IMG_RESHAPE_SIZE: 256
    TEMP_LEN: 32
    FRAME_RATE: 1

  MODEL:
    SINGLE_FRAME: False
    BACKBONE_NAME: ViT
    TUNE_POINT: 0 # [0, 11]
    TEMPORAL_DS_STRATEGY: decode    
    LAST_STRIDE: False
    GENERATE_LFB: False
    NAME: 'ucf_vit-b'
    ENC_LAYERS: 3
    DEC_LAYERS: 3
    D_MODEL: 256
    NHEAD: 8
    DIM_FEEDFORWARD: 2048
    QUERY_NUM: 15
    NORMALIZE_BEFORE: False
    DROPOUT: 0.1
    DS_RATE: 8
    TEMP_LEN: 32
    SAMPLE_RATE: 1
    PRETRAINED: True
    PRETRAIN_BACKBONE_DIR: "../pretrained_models/backbone/ViT-B_K400_16x4.pth"
    POSITION_EMBEDDING: "sine"
    NUM_FEATURE_LEVELS: 4
    NUM_ENC_N_POINTS: 8
    BBOX_EMBED_DIFF_EACH_LAYER: False
    LABEL_SMOOTHING_ALPHA: 0.0

  MATCHER:
    COST_CLASS: 4
    COST_BBOX: 5
    COST_GIOU: 2
    BNY_LOSS: True
    BEFORE: False

  LOSS_COFS:
    MASK_COF: 1
    DICE_COF: 4
    BBOX_COF: 5
    GIOU_COF: 2
    PERSON_COF: 3 
    EOS_COF: 0.5
    WEIGHT: 10
    CLIPS_MAX_NORM: 0.1

  LOG:
    BASE_PATH: '/path/to/save/JHMDB'
    LOG_DIR: 'tb_log'
    SAVE_DIR: 'checkpoints'
    EVAL_DIR: '/path/to/save/JHMDB/eval'
    RES_DIR: CM_ViT-B_{}-{}/res
    EXP_NAME: CM_ViT-B_{}-{}
    DISPLAY_FREQ: 1000